Skip to content
search

Guides

Task-oriented walkthroughs for the things people actually do with sx: building an index, searching it, faceting, vectors, hybrid, SQL, operations, and tuning.

Each guide is built around a job rather than a flag: putting documents into an index, getting them back out by relevance, counting them into facets, searching by vector, and keeping a .sx file healthy in production. They assume you have worked through the quick start.

  • Building an index: the schema, field types, analyzers, batch indexing, updates and deletes, and reclaiming space.
  • Full-text search: the query tree, the compact query string, the JSON DSL, boosts, and sx query.
  • Facets and sorting: doc-values, the request API, sort keys, aggregations, and collapse.
  • Vector search: the dense_vector field, HNSW knobs, kNN queries, quantization, and filtered kNN.
  • Hybrid search: fusing a text query and a kNN query with Reciprocal Rank Fusion.
  • The SQL interface: the built-in SELECT surface, MATCH, bind parameters, and the Go API.
  • Operations: inspect, verify, backup, restore, repair, export, import, vacuum, exit codes, and locking.
  • Performance tuning: the latency budget, durability knobs, segment management, and the bench gate.
Building an index Define a schema, pick field types and analyzers, index in batches, replace and delete documents, and reclaim space. Full-text search Build a query tree, parse the compact query string or the JSON DSL, apply boosts, and run sx query. Facets and sorting Sort by a field, count documents into buckets, and collapse duplicates with the request API and the CLI flags. Vector search Define a dense_vector field, index vectors, run kNN queries, tune the HNSW graph, quantize, and filter. Hybrid search Fuse a text query and a kNN query into one ranked list with Reciprocal Rank Fusion. The SQL interface Query a .sx file with a SELECT surface: MATCH, structured predicates, ORDER BY, bind parameters, and the Go API. Operations Inspect, verify, back up, restore, repair, export, import, and vacuum a .sx file, with exit codes and locking notes. Performance tuning Treat latency as a budget, choose PageSize and Sync, keep segments and tombstones down, and guard against regressions with sx bench.